|Title||SGUARD: A feature-based clustering tool for effective spreadsheet defect detection|
|Authors||Da Li, Huiyan Wang, Chang Xu, Ruiqing Zhang, Shing-Chi Cheung, & Xiaoxing Ma|
|Publication||34th IEEE/ACM International Conference on Automated Software Engineering (ASE 2019)|
Spreadsheets are widely used but subject to various defects. In this paper, we present SGUARD to effectively detect spreadsheet defects.
SGUARD learns spreadsheet features to cluster cells with similar computational semantics, and then refine these clusters to recognize anomalous cells as defects. SGUARD well balances the trade-off between the precision (87.8%) and recall rate (71.9%) in the defect detection, and achieves an F-measure of 0.79, exceeding existing spreadsheet defect detection techniques.
This example illustrates use of SGUARD on a worksheet adapted from the EUSES corpus.
It contains three cell clusters, each of which follows a specific computational semantic, as annotated by three colors (green, orange, and blue).
Five cells contain faulty formulas (defects), namely, D15, D19, F13, F16, and F19, as annotated by red triangles.